Why don’t you tell us how to think about the AGI problem? So far you have spent your time writing about how probability theory, statistics, information theory, logic etc are all useless and how “deep learning” is the solution to everything. It is like someone saying that because there is no experimental proof of string theory, it must be that the entire physics field should be thrown off the bus. It is simply ridiculous and incredibly ignorant. I am not a statistician but I took a few statistics classes in the statistics department (e.g. statistical data mining, which is now called machine learning) and were taught to train multi-layer Neural Networks back in 2007, before the name deep learning was even created, however we weren’t taught any bayesian methods. Just like in physics you have String theorists and Particle theorists, the statisticians are split between the bayesians and the frequentists who hate each other, but no one would ever claim that we should get rid of probability theory (which is actually a branch of mathematics under measure theory). Are you suggesting that maybe we should get rid of integration because the Bayesian statisticians use it? Research in Neural Networks is interdisciplinary in nature and many engineers, psychologists (e.g. Hinton), computer scientists, statisticians, even economists (Werbos) built this field. I haven’t seen your name in any research papers, so what exactly is your contribution to this field you call “Deep Learning”? Failed physicists in particular have a habit of trying to enter other fields by claiming that everyone else is useless, then trying to re-invent the wheel and then subsequently failing to deliver anything useful. Does anyone remember derivative pricing using PDEs, Econophysics etc? If physics has all the answers then tell us, what is dark matter and dark energy and what happens in a black hole?

Deep Learning in Not Probabilistic Induction

Carlos E. Perez

7924